[ADMB Users] mcmc screen output

[Arni Magnusson]

ADMB is hardwired to report every 200th MCMC cycle (lines 618-627 in 
src/nh99/xxmcmc.cpp inside 
http://admb-project.googlecode.com/files/admb-9.1-src.zip.

I just ran a simple benchmark, measuring how many seconds it took to run 
'catage -mcmc 1e5 -mcsave 1e2':

   19.5 in a maximized Dos shell
   18.2 in my normal Dos shell (27 rows)
   18.1 in a minimized Dos shell
   17.5 piping to > mcmc_log.txt
   17.4 piping to > NUL

It turns out that the R overhead is surprisingly small:

   17.9 system("catage -mcmc 1e5 -mcsave 1e2")

When running multiple models, I have used R to write Dos/Bash shell 
script(s) and then run the script(s) in a minimized shell, but altogether 
the speed differences above are smaller than I expected.

At the risk of sounding like a scratched record, I will add that by far 
the most significant speed improvement is running the same model in Linux:

   10.7 in a maximized Bash shell
   10.6 in my normal Bash shell
   10.0 in a minimized Bash shell
    9.8 piping to > mcmc_log.txt
    9.8 piping to > /dev/null
    9.8 in R

This would save many hours in Allan's case. These results agree with 
previous ADMB benchmarks 
(http://admb-project.org/community/benchmarks/compilers).

Arni



On Mon, 28 Jun 2010, Jim Ianelli wrote:

> and optionally, you could make use of cerr vs cout.  I.e., make your own 
> counter and send it to cerr and execute as follows:
>
> mymod >NUL
>
> and the screen will display your cerr statements and dump everything else.
>
>
> Jim Ianelli
> NOT sent from my iphone...
>
>
> On 6/28/2010 3:31 PM, Steve Martell wrote:
>> Use >NUL this wil suspend all screen output
>> 
>> Steve Martell
>> Martell.Steve at gmail.com
>> Sent from my iPhone
>> 
>> 
>> On 2010-06-28, at 2:14 PM, Allan.Hicks at noaa.gov wrote:
>> 
>>> Hi All,
>>> 
>>> Does anyone know how to change the frequency at which mcmc output is
>>> displayed to the screen?  For example, instead of the acceptance rate
>>> reported every 200th mcmc draw, is it possible to have it report every
>>> 1000th?  Screen output can be time consuming and I am trying to reduce
>>> the time of my simulations.
>>> 
>>> The options -nox and -iprint seem to have an effect only during the
>>> estimation portion and not during the mcmc.
>>> 
>>> Thanks,
>>> Allan
>>> _______________________________________________
>>> Users mailing list
>>> Users at admb-project.org
>>> http://lists.admb-project.org/mailman/listinfo/users
>> _______________________________________________
>> Users mailing list
>> Users at admb-project.org
>> http://lists.admb-project.org/mailman/listinfo/users
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users
>